ChatGPT is ready to connect to your bank accounts. Should you let it in?

By Julian Torres

OpenAI is letting users link their bank accounts to its chatbot, but cybersecurity experts have safety concerns. Here's what you should know.

ChatGPT Plus users can now link their financial accounts directly to their chats using Plaid.

You probably wouldn't show a stranger your bank account, but OpenAI is betting you would let a chatbot take a look.

OpenAI is expanding Finances, its personal-finance feature, and has started rolling it out to eligible ChatGPT Plus users in the U.S. With the $20-a-month Plus subscription, users can now connect their financial-account data directly to their chats for real-time, tailored answers to questions about their balances, transaction history, portfolio allocation, tax circumstances and more.

ChatGPT's Finances feature is powered by Plaid, a technology company that connects users' banking and investment accounts to apps such as Rocket Money and Venmo. This gives the AI chatbot up-to-date information on the financial accounts you choose to link to it, so it can tailor its guidance to adapt as you spend, save and invest.

While this feature can only read a user's financial data - meaning ChatGPT cannot actually move your money, pay bills or place trades - it can still track a virtually limitless number of highly personalized financial metrics, including monthly cash flow, net worth and recurring expenses.

It can also take into account "financial memories" provided by the user, such as, "I'm saving up to buy a car early next year," or "I still owe my parents for the loan they gave me," which the AI can use to make future advice more relevant.

The difference in ChatGPT responses with and without the new Finances feature connected.

But is giving ChatGPT access to your finances even safe?

The new feature quickly drew skeptical reactions online from people concerned about privacy. There was similarly mixed feedback when OpenAI first posted on X announcing these capabilities to Pro members last month. X user @ndrewZehnder replied to the post, "What sane individual feels comfortable giving this level of access to OpenAI (or any other company for that matter)."

Rival service Perplexity faced a similar response to the release of their financial tools more than two months ago, with users expressing caution over essentially giving a live feed of their financial accounts to AI giants.

News Corp, owner of MarketWatch, has a content-licensing partnership with OpenAI. In response to a request for comment, OpenAI referred MarketWatch to its product release page.

Cybersecurity experts and certified financial planners, or CFPs, said consumers are right to think twice before readily sharing their personal financial information with AI chatbots. "I would compare the risk of sharing your financial data with ChatGPT to sharing it with Instagram, Facebook, Twitter or similar platforms," Vitaly Shmatikov, professor of computer science at Cornell Tech, told MarketWatch in an email.

ChatGPT Pro and Plus users can now connect their bank accounts and finance apps to the AI chatbot to get more personalized financial advice.

Shmatikov said that users face two main risks: surveillance and their data being leaked. When people give chatbots access to their financial history, it can merge those details with everything else it knows about their physical health, relationships and interests to create a profile that can be used for highly targeted marketing, Shmatikov said.

"Your models end up knowing an enormous amount of information about you," Rosanna Bellini, assistant professor at New York University's Tandon School of Engineering, told MarketWatch. Users can often feed the AI intimate details they might hesitate to tell their banks or partners. When combined with current financial data, the model can infer a level of context about a person's private life that many fail to realize, Bellini explained.

Beyond surveillance, Shmatikov warned that your chats could be "remembered" by the AI when they are used to train the next models. OpenAI's privacy policy states that it "de-identifies" your information, including prompts, to retrain models - but that doesn't prevent data leakage if your prompt contained sensitive information. The AI could unintentionally disclose that information to other users or be accessed in response to a specifically crafted prompt, Shmatikov warned.

That raises serious questions about the way ChatGPT's personal-finance feature is marketed, because it's distinguished by its ability to provide highly tailored guidance based on the amount of context you provide. The line between giving just enough personal information for a better prompt, and dangerous oversharing, can be thin when you're interacting with a chatbot. Another key point: OpenAI's personal-finance feature is not a registered adviser or fiduciary. That means the chatbot is under no obligation to prioritize the user's interests over its own.

"The fiduciary point is the part most people miss, and it sounds like a technicality until it's your data," Jeff Judge, CFP at Chesapeake Financial Planners, told MarketWatch. The risk in these new personal-finance tools is that they're aggregating your financial information in a less regulated system that you don't control. "When I take in client data, I'm bound by SEC privacy rules and a fiduciary duty," Judge said. "A consumer AI tool is bound by its own terms of service, which it can rewrite."

Another potential concern when linking accounts to an AI involves the privacy of others. Users who link joint or custodial accounts to the chatbot may inadvertently reveal confidential financial information about their spouse or children without their consent, Bellini said.

For those who still choose to use the tool, Bellini recommends approaching it with "extreme caution" and suggested taking the following steps to limit exposure:

Limit identifying information

Never input prompts with information that can be tied back to you, even if it's anonymized later. That includes referencing highly specific context on your financial situation, account numbers or login information.

This is especially important when giving AI access to documents. "The convenience is real. So is the tradeoff," Judge said. Inputting something like your mortgage could easily expose data like your address and account information.

Opt out of model training

AI models like ChatGPT frequently use your data to train their future models unless you explicitly tell it not to. You can opt out of model training in settings under "data controls," which allows you to turn off improving the model with your inputs and face less risk of data leakage.

Enable two-factor authentication (2FA)

Setting up 2FA helps block bad actors who might know your password from accessing your private chat history. This small but crucial step can better protect your sensitive financial history.

Regularly monitor and clear 'financial memories'

Frequently review and delete the AI's financial memories. While these can provide the AI with context for more tailored guidance, it increases the risk of an attacker gaining highly personal information about your financial circumstances.

-Julian Torres

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

07-01-26 0944ET